Bayesian Kelly Strategy

Bayesian Kelly Strategy at a glance

Kind of rule	Joining a move (trend following)
Buys or sells	It can buy, and it can bet on falling prices.
Built on	SMA, Standard deviation, variance
Where it was published	TradingView, by its author, as a Pine Script strategy
Tested on	60 cells: 13 coins on 6 chart speeds (5m, 15m, 30m, 1h, 4h, 1D), October 2021 to June 2026
Fee in the test	0.12 percent per side, buys only
Middle result after fees, every cell	−4.9%
Cells above zero	20 of 60
Cells that cleared the quality check	1 of 60
Middle drawdown, every cell	12.9%
Page updated	2026-11-08

Bayesian Kelly Strategy backtest results, every cell

Every cell is in the picture, not only the strongest. To the right of the black line is above zero. The orange line, when you see one, is the middle.

How Bayesian Kelly Strategy works

Rules in this group usually work like this. It waits until price is already moving, then joins that move. It is wrong in small ways many times, and right in a large way a few times. The hard part is the stretch where price chops around and every entry is a fake. The next paragraph is this particular rule, which can be narrower than the group.

It can buy, and it can bet on falling prices. It acts only when several checks agree at once. It leans on SMA, variance; those names are explained on their own pages, further down. The exact line is in the list below, in the author's names, and you can skip that list.

The author did not set a size, so each trade uses whatever amount the code assumes. It holds one trade at a time.

The exact condition, from the code

Same rule, translated from the code, in the author's names. You can skip this list if the paragraph above was enough.

• Enters long when (the bar index mod Rebalance Interval (7)) equals 0 and (((equity × bayes) / the close) − the position size) is above 0.
• Enters short when (the bar index mod Rebalance Interval (7)) equals 0 and not ((((equity × bayes) / the close) − the position size) is above 0) and (((equity × bayes) / the close) − the position size) is below 0.

Buys or sells	It can buy, and it can bet on falling prices.
Tools it calculates	SMA, Standard deviation, variance
A second chart speed	No. It uses one chart speed only.
Only at certain hours	No.
Adding to a trade	It holds one trade at a time.
How much it trades	The author did not set a size, so each trade uses whatever amount the code assumes.
Cost the author assumed	The author assumed no trading cost.

Default settings

The test used the defaults in the middle column. Nothing was turned to help a coin.

Setting	Default	Group
Prior Lookback Window	252
Evidence Lookback Window	60
Rebalance Interval	7
Min Leverage	0
Max Leverage	2
Evidence Weight	0.5
Kelly Fraction	1

How the backtest was run

Every strategy was tried the same way, on the settings the author published.

The years are October 2021 to June 2026. The prices are from Bybit, for 13 widely traded coins, mostly on hourly, 4-hour and daily charts, with some faster charts. That was 47,242 tries. The window stored in the files is 2021-09..2026-06.

599 rules produced a finished try. A rule that finished above zero on at least 12 cells, before any fee, was tried again with a fee of 0.12 percent per side. Only buys are counted in that second try. 273 rules got it.

The method, the aggregate results and a downloadable table are on the study page. What the desk does with the results beyond these pages is its own.

This is the second pass. A fee of 0.12 percent per side is taken on the buy and again on the sell. Only buys are counted. If the rule also bets on falling prices, those trades are not in these numbers.

After fees, 20 of 60 cells finished above zero (33.3%). The median of every cell was −4.9%.

1 of the cells cleared a quality check: at least 50 trades, gains that are not one lucky trade, and a drawdown that is real without taking the whole account.

Finished above zero by chart speed: 15-minute chart, 15m 3 of 12; daily chart, 1D 0 of 11; hourly chart, 1h 9 of 11; 30-minute chart, 30m 6 of 11; 4-hour chart, 4h 2 of 13; 5-minute chart, 5m 0 of 2.

Before fees, it sat 6th on a list sorted by average Calmar. That list charges no fee.

Cells tried after the fee	60
Finished above zero	20 (33.3%)
Middle result, the median of every cell	−4.9%
Middle drawdown, every cell	12.9%
Cells that cleared the quality check	1
Middle result, over the cells that cleared the quality check	+53.4%
Extra, Calmar on the checked rows. Result divided by the deepest fall	5.31
Extra, Sharpe on the checked rows. Result divided by how bumpy the path was	0.02
Middle drawdown, the checked rows	10.1%
Deepest fall, the checked rows	10.1%
Middle number of trades, the checked rows	96

Label: backtest · old prices, buys only, a fee of 0.12 percent per side, every cell · Oct 2021 – Jun 2026. Hypothetical results; see the risk disclosure.

How to read the result

How to read the number

The median of every cell is the result of the rule. The author's settings were not changed to help any coin. The weakest and the strongest cells are not a result on their own.

What the number leaves out

The table also shows +53.4%, taken only over the cells that cleared the quality check. Cells that finished down are left out of that figure, which is why it stands away from the median of every cell. It describes those rows only.

The middle cell closed only 12 trades. With so few trades, the size of the median moves when the years change. How many cells finished above zero is the more settled fact.

1 of 60 cells cleared a quality check: at least 50 trades, a profit factor that is not one lucky trade, and a drawdown that is real without taking the whole account. Profit factor means the gains divided by the losses. Clearing the check describes this sample. It says nothing about the next set of years.

The count above is also split by chart speed. A speed that finished higher, in these years, is one slice of the same rule. It is not a separate strategy.

A place on the no-fee list is not the result after fees. The middle number above is the one after fees.

What this does not show

The useful question, for a rule that joins a move already underway, is what it gave back in the choppy months when there was no trend to join.

Nothing on this page was fit to the cell that finished highest. One coin, one chart speed, or one strong stretch is a single row. It is not the rule. Other years, another exchange, or another fee would be a different measurement.

Questions about Bayesian Kelly Strategy

Words used on this page

Backtest

The rule was tried on old prices. No real money moved.

Cell

One coin on one chart speed. Bitcoin every 4 hours is one cell. Bitcoin once a day is another.

Chart speed

How often a new price bar is drawn. 15m means every 15 minutes, 1h means hourly, 4h means every 4 hours, 1D means once a day.

Fee, per side

A cost taken each time the rule trades. Per side means once on the way in and once on the way out.

Median

Line the cells up from the biggest loss to the biggest gain. The median is the middle one. Half the cells did better and half did worse.

Drawdown

The deepest fall from a high point along the way, as a percent of the account at that high.

Above zero

The account finished with more than it started.

The indicators Bayesian Kelly Strategy uses

Each has its own page: what it measures, how it is read, where it misleads.

• Simple moving average (SMA): the plainest average, and why it still runs half the strategies
• Standard deviation in trading: the volatility measure inside Bollinger Bands and z-scores

"Bayesian Kelly Strategy" was published on TradingView by its author, under the author's own license. This page is an independent reading of those rules and of what they did on these old prices. It is not written with the author, and it is not a recommendation. Not every detail survives the translation onto this page, so read the original before you rely on it.